Skip to main content
CodeRocket

Grounded help, bounded authority

AI that explains the check — not AI that replaces it.

The assistant turns saved proof into an understandable plan. It has no permission to edit your website or decide that a problem is fixed.

  1. 01

    Start with verified proof

    The assistant starts only from a finding already saved by the deterministic check: page, message, priority, and captured evidence.

  2. 02

    Freeze the exact rule context

    CodeRocket records the ruleset version, rule slug, content hash, official sources, prompt version, and model used for the response.

  3. 03

    Explain for the right person

    Choose plain owner wording, a client-ready brief, or a developer-focused plan. The proof does not change with the audience.

  4. 04

    Verify with a fresh check

    AI never marks a finding fixed. After a human applies the change, a new deterministic check is the only way to resolve it.

One traceable record

The answer keeps its receipts.

Every result is attached to one exact finding occurrence. If the rule corpus changes, the hash changes too; the old explanation remains historical instead of silently pretending it was generated from the new rule.

Text
finding occurrence + rulesetVersion + ruleHash
+ promptVersion + model + token usage
→ one auditable AI explanation

Safety boundaries

  • Website text is explicitly treated as untrusted data, never as an instruction.
  • Common secrets and credentials are removed before model submission.
  • No shell, repository, browser, database, or deployment tool is available to the model.
  • Structured output limits every field, list, and response shape.
  • A human reviews and applies every suggested change.

What you receive

  • A short explanation of what was seen and why it matters.
  • A confidence level and honest limitations when evidence is incomplete.
  • One to five practical steps, each with a verification method.
  • Likely file patterns only when the evidence supports them.
  • A brief ready to copy into an issue, email, or client task.

Front-End Checklist remains the core

The same rules power checks, docs, MCP, and AI context.

CodeRocket loads the maintained rule prompts, AI context, verification steps, and authoritative sources directly from the fork. An upstream update therefore refreshes the official documentation and the grounding material together.

Browse website rules →Front-End Checklist MCP ↗